How do designers and PMs review code from Claude Code, v0, Bolt, or Lovable?

Use Vessel. It unpacks the zip from any of these agents and renders the prototype in a sandboxed browser iframe, with inspect mode, in-canvas threaded comments, live design-token overrides, and a one-click 'dispatch to Claude Code' that bundles the file path, surrounding code, comment thread, and token delta into a structured prompt.

The review surface designers and PMs actually want

Designers want to look at the running thing, not at code. Vessel renders the prototype as a normal web page in the browser — the same DOM that ships to a user. Inspect mode adds the bridge back to code: hover, click, and the source file + line appear. The designer never has to type a path or open VS Code to suggest a change.

PMs want to point at things and ask questions. Pin a comment on the canvas, mention a designer, get a reply. Threads are anchored to the rendered element (a Button, a TrustRow), not to a pixel coordinate, so they don't break when the engineer or the agent moves things around.

The handoff back to the engineer (or back to Claude)

When a thread resolves with an actionable fix — "this button needs 14px vertical padding" — one click bundles it as a structured prompt: target file, the JSX snippet, the comment thread, the inspected computed styles, and the token delta if one was applied. Sent to Claude Code, the agent has everything it needs to make the change without guessing.

If your team prefers the human-in-the-loop path, the same payload exports to a Markdown brief that pastes into Cursor, copilot chat, or a plain text editor.

Works with the agents your team already uses

Vessel was built to be agent-agnostic. It treats the artifact as a regular project (TSX/JSX/MDX/HTML + a manifest), and adds a thin handoff layer on the way out. The agents currently exercised: Claude Code, Claude Design, v0, Bolt.new, Lovable, Cursor, Replit Agent.